Do It by Hand, Then Write It Down

The Expensive Shortcut

My support inbox held 212 unread messages on the Monday morning after I automated the delivery of my template catalogue. I had built the system over a long weekend in October 2021, fourteen hours of it, and I went to bed on Sunday pleased with myself. A customer paid, a payment tool told a file-sharing tool, the file-sharing tool sent a download link and a receipt. No more sitting at my desk in Pune at eleven at night, matching orders to files by hand.

I had never written down what I did when I matched them. I didn't think there was anything to write. You see an order, you send the bundle. Simple.

It wasn't simple, and I had been making a decision on every order without noticing. When an order came from America or Canada, I sent the version of the bundle formatted for US Letter paper, with dollar signs in the pricing sheets and "sales tax" in the invoice template. Everyone else got A4, pounds and VAT. I did this by glancing at the billing address, a two-second habit I had picked up in 2019 after an angry email from a bakery in Ohio. Two seconds per order. It was invisible to me, and so it was invisible to the machine.

The automation sent everyone the A4 file. Over nine days, about a hundred and thirty American and Canadian buyers printed invoice templates that cut off at the bottom and asked their clients for VAT they did not owe. Forty of them asked for their money back. At £49 a bundle, that came to £1,960 in refunds, plus the card fees I did not recover, plus roughly twenty-two hours of apologies, replacement files and one long call with a bookkeeper in Calgary who had sent the wrong invoices to eleven of her own customers.

The fourteen hours I spent building the system had saved me about forty minutes a week. On a generous reckoning it would have paid for itself by the following spring. Instead it cost me a month's profit and a fair amount of sleep.

The tools worked perfectly. They did exactly what I had told them, and what I had told them was incomplete, because I did not know my own process well enough to describe it.

The bookkeeper in Calgary wrote to me afterwards. "You must have tested this," she wrote. I had tested it once, with my own address.

Writing It Down Properly

The fix for my 212 unread messages was a document I should have written two years earlier. I call it the Run Log: a plain, running record of every task the business repeats, the steps it takes, the time it costs and the decisions hidden inside it. It is not software. Mine began as a text file called how-i-actually-do-things.txt, and for a long time it stayed one.

The order is fixed: do, document, delegate, automate. You do the task by hand until you understand it, write it down, hand it to someone else, and only then automate the parts that have stopped changing. I skipped the second step and paid for it in Calgary.

A usable entry has six parts, and each one earns its place:

• Task: a short name you would recognise in a list of forty.
• Trigger: the event that starts it.
• Steps: numbered, in the order you actually do them, not the order you wish you did.
• Decisions: every point where you look at something and choose. Mark these. They are where delegation and automation go wrong.
• Time: minutes per step, measured rather than guessed.
• Done when: the single fact that tells you the task is finished.

Meera wrote her first entry the week the Oadby dental surgery signed. Her draft had three steps. When she did the work and wrote down everything she touched, it had twelve.

Task: New client onboarding. Trigger: signed proposal. Done when: first month closed and the client has their deadline calendar.

1. Send the engagement letter and fee schedule; chase until signed.
2. Request photo ID and proof of address from each partner for money-laundering checks.
3. Run the checks. Decision: if any detail fails to match, stop and call.
4. Request read-only access to the bank feed and accounting software.
5. Ask the previous bookkeeper for the handover file.
6. Create the client folder using the standard naming format.
7. Import twelve months of transactions.
8. Reconcile opening balances. Decision: a difference over £50 goes to the client; under £50 goes to suspense with a note.
9. Map the chart of accounts. Decision: split NHS and private income, or not?
10. Check the VAT position. Decision: dental treatment is mostly exempt; refer anything partial to the client's accountant.
11. Agree the monthly cut-off and what the client sends, by which date.
12. Send the welcome email with the deadline calendar; book the first monthly call.

The time column stayed blank. She had no idea yet what any of it cost, and the practice manager's seventeen demands were still pinned above her desk, eleven of them already ticked.

Finding the Hidden Decisions

Four bold decisions out of twelve steps looked about right to Meera. I asked her to go through the list again with one question for each line: what do you look at here, and what could you see that would make you do something different next? Mechanical steps have one answer. You do the thing, and it comes out the same whoever does it. Judgement steps have at least two answers, and the choice depends on something you noticed.

She found three more.

Step 1 said "chase until signed". Chase how often, and for how long? When she thought about it, she chased on day three and day seven, then rang on day ten, and if nothing had happened by day fourteen she quietly let the client go. That last part was a decision she had made four times in two years and never once written down. Step 2 said "each partner". The Oadby practice had two partners and three associate dentists, and she had to decide whether the associates counted. (They didn't, but she had checked with her own accountant the first time, and the answer lived only in her memory.) Step 7 said "import twelve months", which is mechanical until the client changed software in month five and half the year sits somewhere else.

So she rewrote every decision line in the same pattern: what you look at, the rule, the threshold, and who to ask when the rule does not cover it. "Difference over £50" was already in that form. "Split NHS and private income, or not?" was not. It became: split if NHS income is more than 10% of turnover; below that, one income line and a note in the file.

This is the work that makes everything after it possible, and it is dull. A freelancer, a virtual assistant or an automation will do mechanical steps faithfully for ever. They will also do judgement steps mechanically, which means doing them wrong with complete confidence. My system in 2021 did not fail at sending files. It failed at a two-second glance at a billing address that I would have sworn was not a step at all.

Meera's twelve steps became fifteen, with seven decisions. She printed the page and taped it beside the seventeen demands, and the two sheets curled together in the draught from the window that never quite shut.

The Three Bins

The time column filled up the following month, when a physiotherapy clinic in Market Harborough signed. Meera ran the whole onboarding with a kitchen timer on the desk, writing down the minutes beside each of the fifteen steps before she moved to the next. It came to 580 minutes: nine hours and forty, spread over eleven days.

Then she sorted every step into one of three bins:

• Machine: AI or automation can do it reliably, and it comes out the same every time.
• Specialist: a skilled person should do it, but that person need not be you.
• Founder: judgement, relationships and decisions that only you can make.

The rule for sorting is blunt. Put each step in the lowest bin that could do it safely, assuming the written rules are followed. That assumption is why the previous section mattered. Before she wrote "split if NHS income is more than 10% of turnover", step 9 belonged to Meera. Afterwards, a trained bookkeeper could do it.

Six steps went to Machine: the engagement letter, the chasing sequence, the folder, the import, the welcome email and the call booking. Together they took 95 minutes. Six went to Specialist: the identity checks, the handover request, the reconciliation, the account mapping, the VAT check and the deadline calendar. Those took 415 minutes. Three stayed with her: the phone call when an identity check fails, the conversation agreeing what the client sends and by when, and the first monthly call. They took 70 minutes.

Seventy out of 580 is a little over 12 per cent.

"So what exactly am I for?" she wrote to me that evening.

I knew the feeling. When I first sorted my own Run Log in 2021, the Founder bin held 11 per cent of my hours. I spent the weekend moving things back into it, inventing reasons why only I could rename a folder. By Monday I had run out of reasons.

The answer to Meera's question sits in those seventy minutes. That is the part clients actually value. The other 510 are a sequence of known steps with a known cost, and a sequence with a known cost can be priced as a package.

This Week

Choose one task your business repeats at least monthly. Do it twice by hand with a timer running, and write it up in the six parts: task, trigger, steps, decisions, time, done when. Rewrite every decision as a rule with a threshold and a named person to ask. Then sort each step into a bin and add up the minutes in the Founder bin. Write that number on paper, large, where you will see it.

Meera's says 70. It is still taped to the window frame, and the corner has gone soft from being touched.